CN115618215A - Complex electromagnetic environment grading method based on morphological intelligent computation - Google Patents
Complex electromagnetic environment grading method based on morphological intelligent computation Download PDFInfo
- Publication number
- CN115618215A CN115618215A CN202211275076.6A CN202211275076A CN115618215A CN 115618215 A CN115618215 A CN 115618215A CN 202211275076 A CN202211275076 A CN 202211275076A CN 115618215 A CN115618215 A CN 115618215A
- Authority
- CN
- China
- Prior art keywords
- electromagnetic environment
- morphological
- signal
- gradient
- morphology
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R29/00—Arrangements for measuring or indicating electric quantities not covered by groups G01R19/00 - G01R27/00
- G01R29/08—Measuring electromagnetic field characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Electromagnetism (AREA)
- Complex Calculations (AREA)
Abstract
The invention relates to a signal processing technology, in particular to a complex electromagnetic environment grading method based on morphological intelligent computation, which adopts a nonlinear method to process electromagnetic environment signals so as to realize the extraction of electromagnetic environment signal characteristics; the method comprises the steps of classifying complex electromagnetic environment signals, accurately identifying the electromagnetic environment signals with different complexities, providing important basis for post-processing such as signal analysis and interference, accurately and efficiently classifying the electromagnetic environment signals with different complexities, and establishing a calculation model and a training algorithm; a heuristic algorithm is used to initialize the training parameters. Extracting morphological characteristics of the electromagnetic environment signal by a characteristic extraction method based on a logarithmic morphological gradient spectrum; the extracted electromagnetic environment signal characteristic parameters with different complexities have good distinguishability. The electromagnetic environment signal classification precision is high by adopting a calculation model and a training algorithm established by the dendritic morphological neurons.
Description
Technical Field
The invention relates to a signal processing technology, in particular to a complex electromagnetic environment grading method based on morphological intelligent computation.
Background
With the rapid development of wireless communication technology, the system and modulation pattern of communication signals are complex and various, and the frequency spectrum is increasingly crowded and overlapped, so that the background noise and interference are obviously improved, and the electromagnetic environment is extremely complex.
Such a complex electromagnetic environment causes severe electromagnetic noise interference to wireless communication systems in both military and civil fields, and even interrupts communication, thereby placing higher demands and more serious challenges on wireless communication systems, especially on signal detection and estimation at the receiving end.
The complex electromagnetic environment is an electromagnetic environment formed by electromagnetic signals which are distributed in space domain, time domain, frequency domain and energy in various quantities, complex in style, densely overlapped and dynamically overlapped in a certain space. The complex electromagnetic environment signal shows the characteristics of typical nonlinearity, non-stability and strong noise interference. At present, when the electromagnetic environment signal feature extraction is faced, a plurality of problems still exist, and the classification precision of electromagnetic signals is not high.
Disclosure of Invention
The invention aims to provide a complex electromagnetic environment classification method based on morphological intelligent computation to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme:
a complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps:
step one, processing an electromagnetic environment signal by a nonlinear method to realize extraction of electromagnetic environment signal characteristics;
step two, grading the complex electromagnetic environment signals, accurately identifying the electromagnetic environment signals with different complexities, and providing important basis for post-processing such as signal analysis, interference and the like; accurately and efficiently classifying electromagnetic environment signals with different complexities, and establishing a calculation model and a training algorithm;
and step three, initializing the training parameters by using a heuristic algorithm.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: in the first step, the mathematical morphology spectrum algorithm adopted for electromagnetic environment signal feature extraction is a logarithmic morphology gradient spectrum algorithm.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: when the feature of the electromagnetic environment signal is extracted by adopting a logarithmic form gradient spectrum algorithm, firstly, a proper maximum analysis scale parameter and a proper structural element are selected, then, multi-scale expansion and corrosion operations are carried out on the electromagnetic environment signal with different complexity, and the logarithmic form gradient spectrum value of the signal is calculated to be used as the feature vector of the electromagnetic environment signal, so that a feature set of the electromagnetic environment signal based on self-adaptive multi-scale form gradient change and non-negative matrix decomposition, a feature set based on the logarithmic form gradient spectrum and a data set formed by the two are obtained.
The complex electromagnetic environment classification method based on morphological intelligent computation as described above: in the second step, the electromagnetic environment signals with different complexities are accurately identified by selecting effective characteristics capable of representing characteristic differences among the electromagnetic environment signals with different complexities.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: a calculation model and a training algorithm based on a dendritic morphology neural network with a random gradient descent are adopted, and a heuristic algorithm is used for initializing learning parameters.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: the electromagnetic environment signal is classified based on self-adaptive multi-scale morphological gradient change and a nonnegative matrix decomposition feature set, a feature set based on a logarithmic morphological gradient spectrum and a data set formed by the self-adaptive multi-scale morphological gradient change and the nonnegative matrix decomposition feature set through a dendritic morphological neural network computing model based on random gradient descent and a training algorithm.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: in the dendritic morphology neural network based on random gradient descent, dendritic morphology neurons use a super box as a classification decision boundary, and each dendritic neuron generates a super box, wherein the most active dendritic cell is closest to an input class.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: a softmax layer is added at the end of the dendritic morphology neuron, and the dendritic output is normalized and used as a measurement parameter of the class possibility.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: and in the third step, initializing the super box by adopting a heuristic method.
The complex electromagnetic environment grading method based on morphological intelligent computation comprises the following steps: the initialization of the hyper-boxes is using HpC initialization and dHpC initialization.
Compared with the prior art, the invention has the beneficial effects that: extracting morphological characteristics of the electromagnetic environment signal by a characteristic extraction method based on a logarithmic morphological gradient spectrum; the extracted electromagnetic environment signal characteristic parameters with different complexities have good distinguishability.
The electromagnetic environment signal classification precision is high by adopting a calculation model and a training algorithm established by the dendritic morphological neurons.
Drawings
FIG. 1 is a simple electromagnetic environment signal waveform of the present invention.
Fig. 2 is a waveform of a mild complex electromagnetic environment signal in accordance with the present invention.
FIG. 3 is a waveform of a medium complexity electromagnetic environment signal in accordance with the present invention.
Fig. 4 shows the waveform of a heavily complex electromagnetic environment signal in the present invention.
FIG. 5 is a basic mathematical form spectrum of electromagnetic environment signals of four different complexities of linear structural elements in the invention.
FIG. 6 is a basic mathematical form spectrum of four electromagnetic environment signals with different complexities of square structural elements in the invention.
FIG. 7 is a basic mathematical shape spectrogram of electromagnetic environment signals of four different complexities of diamond-shaped structural elements in the invention.
FIG. 8 is a shape gradient spectrum of electromagnetic environment signals of four different complexities of linear structural elements in the present invention.
Fig. 9 is a shape gradient spectrogram of electromagnetic environment signals of four different complexities of square structural elements in the invention.
FIG. 10 is a shape gradient spectrum of electromagnetic environment signals of four different complexities of diamond-shaped structural elements in the present invention.
FIG. 11 is a logarithmic form gradient spectrum of electromagnetic environment signals of four different complexities of linear structural elements in the present invention.
FIG. 12 is a logarithmic shape gradient spectrum of electromagnetic environment signals of four different complexities of square structural elements in the invention.
FIG. 13 is a logarithmic shape gradient spectrum of electromagnetic environment signals of four different complexities of diamond-shaped structural elements in the present invention.
FIG. 14 is a schematic diagram of a neuron according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
As an embodiment of the invention, the complex electromagnetic environment classification method based on morphological intelligent computation comprises the following steps:
and processing the electromagnetic environment signal by adopting a nonlinear method to realize the extraction of the electromagnetic environment signal characteristics.
The complex electromagnetic environment signal shows the characteristics of typical nonlinearity, non-stability and strong noise interference; therefore, the adoption of a nonlinear method to process the electromagnetic environment signal so as to realize the extraction of the electromagnetic environment signal characteristics is the core and key of the evaluation of the environment signal complexity;
in the invention, the adopted nonlinear method is mathematical morphology, the mathematical morphology is a nonlinear analysis method, the morphological characteristics of the signals are described from the perspective of aggregation, the logic is strict, and the algorithm is simple.
The mathematical morphology-derived mathematical morphology spectrum algorithm is widely applied to the fields of fault diagnosis, image processing and the like by utilizing the mathematical morphology spectrum algorithm to extract nonlinear features, but the application of the mathematical morphology-derived mathematical morphology spectrum algorithm to the extraction of the electromagnetic environment signal features needs to be further developed.
Mathematical morphology spectrum algorithm can be used for analyzing the characteristics and structural shapes of images and signals; specifically, structural elements with different scales are adopted to process signals, and structural morphological characteristics of the signals on different scales are reflected.
The mathematical morphology spectrum algorithm comprises a basic mathematical morphology spectrum algorithm, a morphology gradient spectrum algorithm and a logarithmic morphology gradient spectrum algorithm;
and respectively extracting the characteristics of the electromagnetic environment signals by using a basic mathematical morphology spectrum algorithm, a morphology gradient spectrum algorithm and a logarithmic morphology gradient spectrum algorithm, and comparing the results. The result comparison shows that the feature extraction method based on the logarithmic form gradient spectrum algorithm is more comprehensive in extraction of the form features of the electromagnetic environment signals, and the extracted feature parameters have better characterization effects; therefore, the mathematical morphology spectrum algorithm adopted in the invention is preferably a logarithmic morphology gradient spectrum algorithm.
For ease of understanding, the basic mathematical morphology spectrum algorithm, the morphology gradient spectrum algorithm, and the logarithmic morphology gradient spectrum algorithm are explained below, respectively, as follows:
(1) A basic mathematical morphology spectrum algorithm;
the method comprises the following steps of (1) performing a basic mathematical morphology spectrum algorithm, namely mathematical morphology particle analysis, wherein the mathematical morphology particle analysis is a multi-scale analysis method based on mathematical morphology and is widely applied to the fields of image processing and mechanical signal processing;
the mathematical morphology particle analysis specifically comprises the steps of filtering a research object by using structural elements with different scales and different shapes to obtain internal information of the research object;
the mathematical morphology particle analysis is defined by the formula: y = { psi λ } λ≥0 I.e. a series psi λ A collection of transforms; transforming psi λ Is defined as:
in the above formula, the first and second carbon atoms are,
is a morphological open operation symbol, g is a structural element; λ is a scale parameter of the structural element; thus, λ g represents a structural element at a certain scale parameter λ; in an open operation environment, the value of lambda is not less than 0;
it can be seen from the formula that the mathematical morphology particle analysis is based on mathematical morphology open operation.
Wherein, the expression of λ g is as follows:
in the above formula, the first and second carbon atoms are,
is a morphological dilation operator.
The basic mathematical morphology spectrum algorithm takes mathematical morphology particle analysis as an operation basis, a scale parameter lambda is a control variable, and the distribution condition of the morphology particles under different scales is intuitively reflected in a curve form.
If a continuous time domain function is defined as f (n) and a structure function is defined as g (m), the basic mathematical morphology spectrum algorithm function MS (f, λ, g) is expressed as follows:
in the above formula, s =: (x) dx, representing the measure of f (x) within the domain of definition; the gray value signal usually takes the area of the local shadow as the measure; therefore, when λ ≧ 0, the basic mathematical morphology spectrum algorithm corresponding to the function f (n) under the function of the scale structure function λ g is also called as the mathematical morphology spectrum algorithm of division operation, which can be denoted as MS + ；
The mathematical morphology spectrum algorithm corresponding to the open operation mathematical morphology spectrum algorithm is a closed operation mathematical morphology spectrum algorithm and can be recorded as MS-; in the closed operation mathematical morphology spectrum algorithm, lambda is less than 0; the expression of the basic mathematical morphology spectrum algorithm function MS (f, λ, g) is as follows:
in the above formula, "·" is a morphological closed operation symbol.
In the morphological particle analysis of one-dimensional discrete signals, the basic mathematical morphology spectrum algorithm can show stronger sensitivity considering the change of the value of the lambda, so the lambda can be taken as a continuous integer value, and the maximum value of the lambda is the lambda max And minimum value of λ respectively min ；
At this time, the mathematical morphology spectrum algorithm of the on operation and the off operation of the one-dimensional discrete signal can be respectively simplified, and the simplified expressions are as follows:
MS - (λ,g)＝S'[f·(-λ)g-f·(-λ+1)g] λ min ≤λ＜0；
in the above formula, S' = ∑ f (n).
Since the mathematical morphology spectrum algorithm of the open operation has non-expansibility, i.e.
The mathematical shape spectrum algorithm of the open operation of the one-dimensional discrete signal is not necessarily negative, i.e. MS + (lambda, g) is more than or equal to 0; similarly, the closed-operation mathematical morphology spectrum algorithm has the expansibility, i.e. f · (lambda-1) g ≦ f · lambda g, i.e. MS _ (lambda, g) is more than or equal to 0, so that meaningful non-negative spectral lines are ensured in the form spectrogram of the one-dimensional discrete signal. Similar to the meaning of the frequency spectrum in fourier transform, the basic mathematical morphology spectrum reflects the distribution of signal morphology features at different structural element scales. For a structural element λ g under a certain scale, when the signal has more morphological structural components corresponding to it, the spectral line value represented in the mathematical morphological spectrum of the signal is larger, whereas the less structural components correspond to the smaller spectral line value.
The basic mathematical morphology spectrum can be simplified into an open operation mathematical morphology spectrum when the dimension lambda is more than or equal to 0 and a closed operation mathematical morphology spectrum when the dimension lambda is less than 0, wherein the open operation mathematical morphology spectrum reflects the structural characteristic information of the signal, and the closed operation mathematical morphology spectrum reflects the corresponding background information. From the duality of the open and close operations: the open and close operation mathematical morphology spectrums are basically consistent in the aspect of describing the morphology complexity of the signals, and generally, the signal structure characteristic information can be fully embodied only by researching the open operation mathematical morphology spectrums.
(2) A morphological gradient spectrum algorithm;
when morphological dilation and morphological erosion are performed on the time domain function f (n) using the structure function g (m), the difference between the two constitutes the concept of morphological gradient. When the signal is processed, the morphological gradient combines the characteristics of morphological expansion and morphological erosion, the positive pulse information and the negative pulse information of the signal can be considered simultaneously, and the morphological characteristics of the signal can be effectively extracted.
In the morphological gradient spectrum algorithm, the expression of the morphological gradient operator is as follows:
introducing the morphological gradient operator into a basic mathematical morphological spectrum algorithm function to obtain the morphological gradient spectrum algorithm function, wherein the expression is as follows:
for a one-dimensional discrete signal, since the morphological gradient operator has expansibility, the morphological gradient spectrum algorithm function can be simplified, and the simplified function expression is as follows:
in the above formula, MGS + (f,. Lambda.g) and MGS - (f, lambda and g) are respectively an expansion morphological gradient spectrum algorithm function and a corrosion morphological gradient spectrum algorithm function; the two respectively reflect the shape change rule of the signal under different structural element scales in the positive and negative intervals. From the duality of the dilation and erosion operations, it can be seen that the two morphology gradient spectra are essentially the same structure in describing the morphological complexity of the object.
(3) A logarithmic morphological gradient spectrum algorithm;
although both the basic mathematical morphology spectrum algorithm and the morphology gradient spectrum algorithm describe the shape change rule of the signal on different scales, the basic mathematical morphology spectrum algorithm has statistical deviation, and the morphology gradient spectrum algorithm mainly highlights the pulse information of the signal. For electromagnetic environment signals, the discrimination between electromagnetic environment signal spectral lines with different complexities is found to be poor by extracting basic mathematical morphology spectral features of the electromagnetic environment signals under different scales; the form gradient spectrum of the electromagnetic environment signals is calculated under different scales, and the obtained curve has certain distinguishing capability on the electromagnetic environment signals with different complexities, but the distinguishing effect is not good.
In order to solve the problems, a logarithmic form gradient spectrum algorithm is introduced, and the complex electromagnetic environment signal characteristics based on the logarithmic form gradient spectrum are extracted by carrying out logarithmic processing on the form gradient spectrum.
Because of the expansion form gradient spectrum algorithm function MGS on the description of the complexity of the signal form + (f, lambda, g) and corrosion morphology gradient spectrum algorithm function MGS - Since (f, λ, g) has consistency, only the expansion form gradient spectrum of the signal needs to be considered in terms of reflecting the structural feature information of the signal.
By applying a function MGS to the gradient spectrum algorithm of the swelling shape + (f, lambda, g) logarithmic processing is carried out, and an expression of the logarithm form gradient spectrum algorithm function on the positive region can be obtained, wherein the expression is as follows:
LMGS + ＝log(MGS + (f,λ,g)+1) λ≥0；
LMGS in the above formula + A log-morphological gradient spectral algorithm function.
According to the logarithmic form gradient spectrum algorithm function on the positive region described by the formula, the complex electromagnetic environment signal characteristics can be extracted; the method comprises the steps of firstly selecting proper maximum analysis scale parameters and structural elements, then carrying out multi-scale expansion and corrosion operation on electromagnetic environment signals with different complexities, calculating a logarithmic form gradient spectrum value of the signal according to the formula to be used as a feature vector of the electromagnetic environment signals, and obtaining the electromagnetic environment signals based on self-adaptive multi-scale form gradient change and non-negative matrix decomposition feature set, the feature set based on the logarithmic form gradient spectrum and a data set formed by the two.
The key of extracting the electromagnetic environment signal characteristics by using a mathematical morphology spectrum algorithm is that whether the obtained signal characteristic parameters can comprehensively reflect the signal structure characteristic information and can stably and effectively distinguish the electromagnetic environment signals with different complexities;
therefore, four electromagnetic environment signals with different complexity are simulated for analysis; the sampling frequency is selected to be f =20000MHz, the sampling time is 2 mus, and the rationality and the effectiveness of the extraction of the electromagnetic environment signal characteristics based on the logarithmic form gradient spectrum algorithm are further verified and analyzed in the Matlab environment.
Time domain waveforms of four electromagnetic environment signals with different complexities are shown in figures 1-4.
And respectively calculating a basic mathematical form spectrum MS, a form gradient spectrum MGS and an logarithm form gradient spectrum LMGS of the electromagnetic environment signals with four different complexities by using a basic mathematical form spectrum algorithm, a form gradient spectrum algorithm and an logarithm form gradient spectrum algorithm, carrying out comparative analysis, and respectively selecting three structural elements, namely a linear structure, a square structure and a rhombic structure in the calculation process.
Basic mathematical form spectrograms of electromagnetic environment signals of four different complexities of three structural elements of linear, square and rhombus can be seen in fig. 5-7; the morphology gradient spectrogram can be seen in figures 8-10; the logarithmic morphological gradient spectrum can be seen in FIGS. 11-13.
By comparing fig. 5 to fig. 7, it can be found that different structural elements exhibit similar processing effects on the obtained basic mathematical morphology spectrum MS curve; the method shows that when the time domain signal of the electromagnetic environment signal is subjected to basic mathematical morphology spectrum characteristic extraction, linear, square and diamond structural elements have similar influences on the time domain signal.
The same analysis of the curves of fig. 8 to 13 can result in similar effects of linear, square and diamond-shaped structural elements on the morphological gradient spectrum and the logarithmic morphological gradient spectrum.
As can be seen from FIG. 11, the log-morphological gradient spectrum LMGS curves of four electromagnetic environment signals with different complexities have good distinctiveness, the spectrum value of each curve monotonically decreases with the increase of the scale parameter of the structural element, and the change trends are consistent. And with the continuous reduction of the spectrum value, the difference value of the spectrum curve of each environment signal is not obviously reduced, which shows that the feature extraction algorithm based on the log morphology gradient spectrum LMGS has good stability.
In fig. 5, although the MS curve of the basic mathematical morphology spectrum shows a decreasing trend as a whole, the MS curve of the basic mathematical morphology spectrum does not decrease steadily with increasing scale parameters, the fluctuation of the spectrum values is obvious, and after the scale parameters are greater than 15, different signal spectrums begin to interleave with each other, so that it can be seen that electromagnetic environment signals with different complexities cannot be distinguished when the MS values calculated by the basic mathematical morphology spectrum algorithm are used as feature values.
In fig. 8 to 10, the trend of the MGS curves of the four morphological gradient spectrums is consistent and monotonously decreasing, and there is a certain degree of distinction when the scale is small, but the degree of distinction of the four curves decreases with the increase of the scale.
Therefore, the characteristic extraction effect on the electromagnetic environment signals with different complexities is good through the logarithmic form gradient spectrum algorithm, and the characteristic parameters of the electromagnetic environment signals with different complexities extracted by the characteristic extraction effect have good distinguishability.
In the method, the complex electromagnetic environment signals need to be classified, the electromagnetic environment signals with different complexities are accurately identified, and important bases are provided for post-processing of signal analysis, interference and the like; and electromagnetic environment signals with different complexities are accurately and efficiently classified.
The method comprises the steps of selecting effective characteristics capable of representing characteristic differences among electromagnetic environment signals with different complexities, and accurately identifying the electromagnetic environment signals with different complexities.
Extracting the characteristics of electromagnetic environment signals with different complexities through a logarithmic form gradient spectrum algorithm;
accurately identifying the extracted features of the electromagnetic environment signals with different complexities, and selecting and designing a classifier to intelligently classify the accurately identified features.
For this purpose, the invention applies dendritic morphological neural networks to the identification of electromagnetic environment signals.
Wherein the dendritic morphology neural network comprises a plurality of dendritic morphology neurons, a dendritic morphology neuron is an artificial neural network using minimum and maximum operators rather than algebraic products; in dendritic morphology neurons, morphological operators construct hyper-boxes in N-dimensional space; these hyper-boxes establish heuristic based training methods without using optimization methods;
because of adopting the training method based on the heuristic method, the convergence problem does not exist, the perfect classification can be achieved, and the training speed is very high. However, perfect classification is inconvenient in practical applications because it can lead to overfitting of the training data, increasing the complexity of the learning model;
in order to solve the problems, the invention adopts a calculation model and a training algorithm based on a dendritic morphology neural network with a descending random gradient, applies the model to classification of electromagnetic environment signals with different complexities, and initializes learning parameters only by using the heuristic algorithms.
In a traditional artificial neural network, the output of a neuron is obtained by performing linear operation of addition, subtraction, multiplication and division on the input of the neuron;
in a conventional artificial neural network, a calculation formula of a neuron is as follows:
in the above formula, τ j Represents the output of the jth neuron, x i Is a real number set and represents the value of the ith neuron connected with the jth neuron; w is a ij Representing the connection weight of two neurons, θ j Representing the activation threshold of the jth neuron.
The schematic diagram of the neuron is shown in fig. 14.
Different from the traditional artificial neural network, the dendritic morphology neuron is a nonlinear neuron, does not use the linear algorithms of addition, subtraction, multiplication and division and the like in the traditional artificial neural network, and replaces the linear algorithms with a lattice index system, thereby realizing the combination of the mathematical morphology and the traditional artificial neural network.
Dendritic morphological neurons have distinct dendritic postsynaptic regions that receive axonal-terminal branch input from other neurons, and the postsynaptic membranes of the dendrites respond with stimulation or inhibition to the input signals they receive. Neurons respond to total dendritic input and are activated by an activation function.
Dendritic morphology neurons use the super-box as their classification decision boundary, and each dendrite will generate a super-box, where the most active dendrite is closest to the input class.
In the scope of classical perceptrons, the most important training method is gradient descent, and has been applied to training multi-layer perceptrons, developing back propagation algorithms; most of the training methods of the morphological neurons are based on intuition rather than gradient descent;
although the existing morphological neuron training method is also based on gradient descent, the method is only suitable for regression tasks.
Due to morphological non-difference, the extension of the above method in classification cannot be directly applied;
for this purpose, the invention adds a softmax layer at the end of the dendritic morphology neuron; thereby deriving a dendritic morphology neuron based on random gradient descent training; the softmax layer is a flexible maximum value transmission function layer and is a category of a neural network classification layer.
In the invention, dendritic outputs are normalized by adding a softmax layer at the tail end of the dendritic morphological neuron, so that the outputs are limited to be between 0 and 1; these outputs may be used as a metric of the likelihood of the class, so that it can be determined to which class of the class the input vector belongs.
In a conventional dendritic morphology neuron, one superbox is represented by its extreme point; in the present invention, the minimum pole and the size vector thereof are used to represent the minimum pole.
In the training of the morphological neural network, the number of the super boxes and the dendritic parameters are automatically determined. This is very different from conventional sensors, where only the learning parameters are determined during the training process.
In contrast, the present invention initializes the hyper-boxes based on heuristic algorithms and optimizes the dendrite parameters using stochastic gradient descent based.
For initializing the hyper-box using a heuristic-based approach, hpC initialization and dHpC initialization methods are used.
In the HpC initialization method, each training set of the same class is contained in a super-box; the maximum number of dendrites is equal to the number of classes.
The dHpC initialization method is used to partition each superbox generated by HpC.
In the invention, a logarithmic form gradient spectrum algorithm is adopted to carry out feature extraction on an electromagnetic environment signal sample to obtain an electromagnetic environment signal feature set based on self-adaptive multi-scale form gradient change and non-negative matrix decomposition feature set, a feature set based on a logarithmic form gradient spectrum and a data set consisting of the two;
then, the three characteristic parameter sets are classified by using a dendritic morphology neuron based on random gradient descent training;
and finally, initializing the dendritic morphology neurons trained on the random gradient descent.
The specific classification results are as follows:
500 samples are collected for each complex electromagnetic environment signal, for a total of 2000 samples. When the DMN-SGD is adopted to carry out complexity identification and classification on electromagnetic environment signals, 400 sample features are randomly selected from the electromagnetic environment signals with each complexity to serve as training samples, the remaining 100 samples are used as testing samples, and in order to obtain accurate and stable experimental results, the obtained experimental data are the average value of 20 classification results.
The DMN-SGD is respectively initialized by adopting K-means, hpC and dHpC and is analyzed and compared with a Support Vector Machine (SVM) and a multi-layer perceptron (MLP).
TABLE 1 electromagnetic Environment Signal Classification results
Table 1 shows the classification results of 5 classifiers. As can be seen from the data in the table, when the complexity classification is carried out on the electromagnetic environment signals, the DMN-SGD model provided by the text can achieve high classification accuracy. With different initialization methods, the classification result will also change accordingly. But overall, the classification precision of the DMN-SGD is higher than that of the SVM and the MLP.
The above embodiments are exemplary rather than limiting, and embodiments of the invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof.
Claims (10)
1. A complex electromagnetic environment grading method based on morphological intelligent computation is characterized by comprising the following steps:
step one, processing an electromagnetic environment signal by a nonlinear method to realize extraction of electromagnetic environment signal characteristics;
step two, grading the complex electromagnetic environment signals, accurately identifying the electromagnetic environment signals with different complexities, and providing important basis for post-processing such as signal analysis, interference and the like; accurately and efficiently classifying electromagnetic environment signals with different complexities, and establishing a calculation model and a training algorithm;
and step three, initializing the training parameters by using a heuristic algorithm.
2. The method for grading a complex electromagnetic environment based on morphological intelligent computation as claimed in claim 1, wherein in the first step, the mathematical morphology spectrum algorithm used for electromagnetic environment signal feature extraction is a logarithmic morphology gradient spectrum algorithm.
3. The method for grading a complex electromagnetic environment based on morphological intelligent computation of claim 2, characterized in that when extracting the electromagnetic environment signal features by using a log-morphological gradient spectrum algorithm, a suitable maximum analysis scale parameter and structural elements are selected first, then multi-scale expansion and corrosion operations are performed on the electromagnetic environment signals of different complexity, and the log-morphological gradient spectrum value of the signal is calculated as the feature vector of the electromagnetic environment signal, so as to obtain the electromagnetic environment signal feature set based on adaptive multi-scale morphological gradient change and non-negative matrix decomposition feature set, the feature set based on the log-morphological gradient spectrum, and the data set composed of the two.
4. The method for grading a complex electromagnetic environment based on morphological intelligent computation of claim 1, wherein in the second step, the electromagnetic environment signals with different complexities are accurately identified by selecting effective features that can characterize the feature differences between the electromagnetic environment signals with different complexities.
5. The method for grading complex electromagnetic environments based on morphological intelligent computation of claim 4, characterized in that a heuristic algorithm is used to initialize the learning parameters by using a computation model and a training algorithm based on a dendritic morphology neural network with descending random gradient.
6. The complex electromagnetic environment classification method based on morphology intelligent computation of claim 2 is characterized in that electromagnetic environment signals are classified based on self-adaptive multi-scale morphology gradient change and a non-negative matrix decomposition characteristic set, a logarithmic morphology gradient spectrum-based characteristic set and a data set formed by the self-adaptive multi-scale morphology gradient change and the non-negative matrix decomposition characteristic set through a computation model based on a dendritic morphology neural network with descending random gradient and a training algorithm.
7. The method for grading a complex electromagnetic environment based on morphological intelligence computation of any of claims 5 or 6, wherein in the dendrite morphological neural network based on stochastic gradient descent, the dendrite morphological neurons use the super-box as their classification decision boundary, and each dendrite generates a super-box, wherein the most active dendrite is closest to the input class.
8. The method for grading complex electromagnetic environments based on morphological intelligence computation of claim 7, wherein a softmax layer is added at the end of the dendritic morphological neurons, and the dendritic outputs are normalized and used as a measure of the likelihood of class.
9. The method for grading complex electromagnetic environments based on morphological intelligence computation of claim 7, wherein in the third step, a heuristic method is used to initialize the super-box.
10. The method for classifying complex electromagnetic environments based on morphological intelligent computation of claim 9, wherein the initialization of the hyper-box employs HpC initialization and dHpC initialization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211275076.6A CN115618215B (en) | 2022-10-18 | 2022-10-18 | Complex electromagnetic environment grading method based on morphological intelligent computation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211275076.6A CN115618215B (en) | 2022-10-18 | 2022-10-18 | Complex electromagnetic environment grading method based on morphological intelligent computation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115618215A true CN115618215A (en) | 2023-01-17 |
| CN115618215B CN115618215B (en) | 2023-09-01 |
Family
ID=84863385
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211275076.6A Active CN115618215B (en) | 2022-10-18 | 2022-10-18 | Complex electromagnetic environment grading method based on morphological intelligent computation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115618215B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738128A (en) * | 2020-06-17 | 2020-10-02 | 山东卓文信息科技有限公司 | Series fault arc detection method based on morphological filtering and MMG |
| CN114217286A (en) * | 2021-12-10 | 2022-03-22 | 中国人民解放军63893部队 | Radar unintentional modulation signal extraction method based on form component |
-
2022
- 2022-10-18 CN CN202211275076.6A patent/CN115618215B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738128A (en) * | 2020-06-17 | 2020-10-02 | 山东卓文信息科技有限公司 | Series fault arc detection method based on morphological filtering and MMG |
| CN114217286A (en) * | 2021-12-10 | 2022-03-22 | 中国人民解放军63893部队 | Radar unintentional modulation signal extraction method based on form component |
Non-Patent Citations (2)
| Title |
|---|
| HUA-CHEN XI 等: "Feature Extraction for Evaluating the Complexity of Electromagnetic Environment Based on Adaptive Multiscale Morphological Gradient and Nonnegative Matrix Factorization", 《JOURNAL OF ELECTRICAL AND COMPUTER ENGINEERING》, pages 1 - 11 * |
| HUA-CHEN XI 等: "Photovoltaic Power Station Electromagnetic Environment Complexity Evaluation Utilizing Logarithmic Morphological Gradient Spectrum", 《FRONTIERS IN ENERGY RESEARCH》, pages 1 - 8 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115618215B (en) | 2023-09-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Patil et al. | White blood cells image classification using deep learning with canonical correlation analysis | |
| CN111160139B (en) | Electrocardiosignal processing method and device and terminal equipment | |
| Bai et al. | Time series classification based on multi-feature dictionary representation and ensemble learning | |
| Nejad et al. | A new enhanced learning approach to automatic image classification based on Salp Swarm Algorithm | |
| Bodyanskiy | Computational intelligence techniques for data analysis | |
| Li et al. | Automatic modulation classification based on bispectrum and CNN | |
| Brenner et al. | Tractable dendritic RNNs for reconstructing nonlinear dynamical systems | |
| Kirichenko et al. | Two approaches to machine learning classification of time series based on recurrence plots | |
| Urgun et al. | Composite system reliability analysis using deep learning enhanced by transfer learning | |
| Tehrani et al. | Frequency-based multi task learning with attention mechanism for fault detection in power systems | |
| CN113052099B (en) | SSVEP classification method based on convolutional neural network | |
| Chinmayi et al. | Emotion Classification Using Deep Learning | |
| Parker et al. | Nonlinear time series classification using bispectrum‐based deep convolutional neural networks | |
| CN115618215A (en) | Complex electromagnetic environment grading method based on morphological intelligent computation | |
| Kirichenko et al. | Applying recurrence plots to classify time series | |
| Xu et al. | Individual identification of electronic equipment based on electromagnetic fingerprint characteristics | |
| Basavaraj et al. | Reliability-driven time series data analysis in multiple-level deep Learning methods utilizing soft computing methods | |
| Long et al. | A statistical description of neural ensemble dynamics | |
| Dani et al. | Survey on the use of CNN and Deep Learning in Image Classification | |
| Purnamasari et al. | EEG based emotion recognition system induced by video music using a wavelet feature vectors and an artificial neural networks | |
| Pradeep et al. | Identifying Stochasticity in Time-Series with Autoencoder-Based Content-aware 2D Representation: Application to Black Hole Data | |
| Kuila et al. | Feature Extraction and Classification of ECG Signals Through Dimension Reduction | |
| Li et al. | Performance Evaluation System Based on Multi-Indicators for Signal Recognition | |
| Nwaigwe et al. | Graph-based methods coupled with specific distributional distances for adversarial attack detection | |
| Ali et al. | Supervised Machine Learning Algorithms: Classification and Comparison Based on Experimental Analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |